probabilities contain NaN in numpy.random.mtrand.RandomState.choice

[mcrewcow]

Hi, thank you for the package developed!

I have an issue running the tool on my dataset. It was mainly built in Seurat and later converted to .h5ad with SeuratDisk. First there was an error about the _index column, fixed with:

fetal_total.__dict__['_raw'].__dict__['_var'] = fetal_total.__dict__['_raw'].__dict__['_var'].rename(columns={'_index': 'features'})

Now, with

python eval_single_anndata.py --adata_path=/mnt/c/Bioinf/HUMAN_FETAL_RETINA/COMBINED_EKPB_v1_clean_int_indexed.h5ad --dir=/mnt/c/Bioinf/HUMAN_FETAL_RETINA/ --species=human

I receive the following error. The output is provided:

[2024-03-30 03:33:45,762] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Using sample 4 layer model
Proccessing COMBINED_EKPB_v1_clean_int_indexed
10.0
COMBINED_EKPB_v1_clean_int_indexed (113073, 1861)
Wrote Shapes Dict
1861
Max Code: 612
Loaded model:
./model_files/4layer_model.torch
  0%|                                                                                                                                                     | 0/4523 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "eval_single_anndata.py", line 155, in <module>
    main(args, accelerator)
  File "eval_single_anndata.py", line 85, in main
    processor.run_evaluation()
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/evaluate.py", line 146, in run_evaluation
    self.starts_path, shapes_dict, self.accelerator, self.args)
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/evaluate.py", line 235, in run_eval
    for batch in pbar:
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/tqdm/std.py", line 1195, in __iter__
    for obj in iterable:
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/accelerate/data_loader.py", line 377, in __iter__
    current_batch = next(dataloader_iter)
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 628, in __next__
    data = self._next_data()
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 671, in _next_data
    data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 58, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 58, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/eval_data.py", line 65, in __getitem__
    dataset_to_starts=self.dataset_to_starts)
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/eval_data.py", line 128, in sample_cell_sentences
    replace=True)
  File "mtrand.pyx", line 935, in numpy.random.mtrand.RandomState.choice
ValueError: probabilities contain NaN

Thank you for your help!

[Yanay1]

This error usually happens when you have cells with no genes expressed. Please double check that the .X slot contains count values, and that all cells have gene expression.

[mcrewcow]

Hi Yanay1,

I checked the object, it does have both .X and .raw.X slots. I can do all the downstream analysis upon h5seurat to h5ad conversion in scanpy, scvelo, scenic+, etc. I have also noticed that if I convert the 'integrated' assay of Seurat object to .h5ad, then it is the error I described above. Yet if I convert the 'RNA' assay, I get the following:

python eval_single_anndata.py --adata_path=/mnt/c/Bioinf/HUMAN_FETAL_RETINA/COMBINED_EKPB_v1_clean_RNA_indexed.h5ad --dir=/mnt/c/Bioinf/HUMAN_FETAL_RETINA/ --species=human --multi_gpu=True
[2024-04-01 15:54:41,230] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Using sample 4 layer model
Proccessing COMBINED_EKPB_v1_clean_RNA_indexed
Killed

[mcrewcow]

[Yanay1]

What is the result of

min(np.sum(fetal_int.X, axis=0))

Try deleting all the intermediate files created by UCE and then re running

[mcrewcow]

Unfortunately, deleting the intermediate did not help.

This is the output of the command: -2559.4630599647617

[Yanay1]

You cannot have negative numbers in .X. The expression values are used as probability weights. They should be count values.

[mcrewcow]

Oh, I have found that it is the issue of SeuratDisk conversion. The counts are written in .raw.X. So I have transferred them to .X now, it looks like real counts. I have also filtered the genes with min_counts = 40. Still, I keep getting

python eval_single_anndata.py --adata_path=/mnt/c/Bioinf/HUMAN_FETAL_RETINA/COMBINED_EKPB_v1_clean_countsonly_new_indexed_int_maybe.h5ad --dir=/mnt/c/Bioinf/HUMAN_FETAL
_RETINA/ --species=human
[2024-04-01 18:16:22,542] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Using sample 4 layer model
Proccessing COMBINED_EKPB_v1_clean_countsonly_new_indexed_int_maybe
7.616953
COMBINED_EKPB_v1_clean_countsonly_new_indexed_int_maybe (113073, 1861)
Wrote Shapes Dict
1861
Max Code: 612
Loaded model:
./model_files/4layer_model.torch
  0%|                                                                                          | 0/4523 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "eval_single_anndata.py", line 155, in <module>
    main(args, accelerator)
  File "eval_single_anndata.py", line 85, in main
    processor.run_evaluation()
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/evaluate.py", line 146, in run_evaluation
    self.starts_path, shapes_dict, self.accelerator, self.args)
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/evaluate.py", line 235, in run_eval
    for batch in pbar:
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/tqdm/std.py", line 1195, in __iter__
    for obj in iterable:
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/accelerate/data_loader.py", line 377, in __iter__
    current_batch = next(dataloader_iter)
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 628, in __next__
    data = self._next_data()
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 671, in _next_data
    data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 58, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/home/mcrewcow/anaconda3/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 58, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/eval_data.py", line 65, in __getitem__
    dataset_to_starts=self.dataset_to_starts)
  File "/mnt/c/Users/rodri/Downloads/UCE-main/UCE-main/eval_data.py", line 128, in sample_cell_sentences
    replace=True)
  File "mtrand.pyx", line 935, in numpy.random.mtrand.RandomState.choice
ValueError: probabilities contain NaN

[Yanay1]

I am not sure what the error could be besides some cell either containing negative numbers, zero counts, or NaNs. If you want, you can email me a copy of the anndata and I can inspect it. My email is (first name) @ stanford.edu.

It seems the issue happens in the first batch.

Thanks!